DMVCCM.html

   1 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"
   2                "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
   3 <html xmlns="http://www.w3.org/1999/xhtml"
   4 lang="en" xml:lang="en">
   5 <head>
   6 <title>DMV/CCM &ndash; todo-list / progress</title>
   7 <meta http-equiv="Content-Type" content="text/html;charset=utf-8"/>
   8 <meta name="generator" content="Org-mode"/>
   9 <meta name="generated" content="2008/06/08 12:43:26"/>
  10 <meta name="author" content="Kevin Brubeck Unhammer"/>
  11 <link rel="stylesheet" type="text/css" href="http://www.student.uib.no/~kun041/org.css">
  12 </head><body>
  13 <h1 class="title">DMV/CCM &ndash; todo-list / progress</h1>
  14
  15
  16 <div id="table-of-contents">
  17 <h2>Table of Contents</h2>
  18 <div id="text-table-of-contents">
  19 <ul>
  20 <li><a href="#sec-1">1 dmvccm report and project</a></li>
  21 <li><a href="#sec-2">2 [#A] P_STOP and P_CHOOSE for IO/EM</a></li>
  22 <li><a href="#sec-3">3 Initialization   </a></li>
  23 <li><a href="#sec-4">4 [#C] Deferred</a></li>
  24 <li><a href="#sec-5">5 Adjacency and combining it with inner()</a></li>
  25 <li><a href="#sec-6">6 Expectation Maximation in IO/DMV-terms</a></li>
  26 <li><a href="#sec-7">7 Python-stuff</a></li>
  27 <li><a href="#sec-8">8 Git</a></li>
  28 </ul>
  29 </div>
  30 </div>
  31
  32 <div id="outline-container-1" class="outline-2">
  33 <h2 id="sec-1">1 dmvccm report and project</h2>
  34 <div id="text-1">
  35
  36 <p><span class="timestamp-kwd">DEADLINE: </span> <span class="timestamp">2008-06-30 Mon</span><br/>
  37 But absolute, extended, really-quite-dead-now deadline: August
  38 31&hellip;
  39 </p><ul>
  40 <li>
  41 <a href="src/dmv.py">dmv.py</a>
  42 </li>
  43 <li>
  44 <a href="src/io.py">io.py</a>
  45 </li>
  46 <li>
  47 <a href="src/harmonic.py">harmonic.py</a>
  48
  49
  50 </li>
  51 </ul>
  52 </div>
  53
  54 </div>
  55
  56 <div id="outline-container-2" class="outline-2">
  57 <h2 id="sec-2">2 <span class="todo">TODO</span> [#A] P_STOP and P_CHOOSE for IO/EM</h2>
  58 <div id="text-2">
  59
  60 <p><a href="src/dmv.py">dmv-P_STOP</a>
  61 Remember: The P<sub>STOP</sub> formula is upside-down (left-to-right also).
  62 (In the article..not the thesis)
  63 </p>
  64 <p>
  65 Remember: Initialization makes some "short-cut" rules, these will also
  66 have to be updated along with the other P<sub>STOP</sub> updates:
  67 </p><ul>
  68 <li>
  69 b[(NOBAR, n<sub>h</sub>), 'h'] = 1.0       # always
  70 </li>
  71 <li>
  72 b[(RBAR, n<sub>h</sub>), 'h'] = h_.probA  # h_ is RBAR stop rule
  73 </li>
  74 <li>
  75 b[(LRBAR, n<sub>h</sub>), 'h'] = h_.probA * _ h_.probA
  76
  77 </li>
  78 <li id="sec-2.1">P_STOP formulas for various dir and adj:<br/>
  79 Assuming this:
  80 <ul>
  81 <li>
  82 P<sub>STOP</sub>(STOP|h,L,non_adj) = &sum;<sub>corpus</sub> &sum;<sub>s&lt;loc(h)</sub> &sum;<sub>t</sub>
  83 inner(s,t,_h_, &hellip;) / &sum;<sub>corpus</sub> &sum;<sub>s&lt;loc(h)</sub> &sum;<sub>t</sub>
  84 inner(s,t,h_, &hellip;)
  85 </li>
  86 <li>
  87 P<sub>STOP</sub>(STOP|h,L,adj) = &sum;<sub>corpus</sub> &sum;<sub>s=loc(h)</sub> &sum;<sub>t</sub>
  88 inner(s,t,_h_, &hellip;) / &sum;<sub>corpus</sub> &sum;<sub>s=loc(h)</sub> &sum;<sub>t</sub>
  89 inner(s,t,h_, &hellip;)
  90 </li>
  91 <li>
  92 P<sub>STOP</sub>(STOP|h,R,non_adj) = &sum;<sub>corpus</sub> &sum;<sub>s</sub> &sum;<sub>t&gt;loc(h)</sub>
  93 inner(s,t,_h_, &hellip;) / &sum;<sub>corpus</sub> &sum;<sub>s</sub> &sum;<sub>t&gt;loc(h)</sub>
  94 inner(s,t,h_, &hellip;)
  95 </li>
  96 <li>
  97 P<sub>STOP</sub>(STOP|h,R,adj) = &sum;<sub>corpus</sub> &sum;<sub>s</sub> &sum;<sub>t=loc(h)</sub>
  98 inner(s,t,_h_, &hellip;) / &sum;<sub>corpus</sub> &sum;<sub>s</sub> &sum;<sub>t=loc(h)</sub>
  99 inner(s,t,h_, &hellip;)
 100
 101 </li>
 102 </ul>
 103
 104 <p>(And P<sub>STOP</sub>(-STOP|&hellip;) = 1 - P<sub>STOP</sub>(STOP|&hellip;) )
 105 </p></li>
 106 <li id="sec-2.2">P_CHOOSE formula:<br/>
 107 Assuming this:
 108 <ul>
 109 <li>
 110 P<sub>CHOOSE</sub>(a|h,L) = &sum;<sub>corpus</sub> &sum;<sub>s&lt;loc(h)</sub> &sum;<sub>t=loc(h)</sub> &sum;<sub>r&lt;t</sub>
 111 inner(s,r,_a_, &hellip;) / &sum;<sub>corpus</sub> &sum;<sub>s&lt;loc(h)</sub> &sum;<sub>t=loc(h)</sub>
 112 inner(s,t,h_,&hellip;)
 113 <ul>
 114 <li>
 115 or t &gt;= loc(h)?
 116 </li>
 117 </ul>
 118 </li>
 119 <li>
 120 P<sub>CHOOSE</sub>(a|h,R) = &sum;<sub>corpus</sub> &sum;<sub>s=loc(h)</sub> &sum;<sub>r&gt;s</sub>
 121 inner(s,r,_a_,&hellip;) / &sum;<sub>corpus</sub> &sum;<sub>s=loc(h)</sub> &sum;<sub>t</sub>
 122 inner(s,t,h, &hellip;)
 123 <ul>
 124 <li>
 125 or s =&gt; loc(h)?
 126
 127
 128 </li>
 129 </ul>
 130 </li>
 131 </ul>
 132 </li>
 133 </ul>
 134 </div>
 135
 136 </div>
 137
 138 <div id="outline-container-3" class="outline-2">
 139 <h2 id="sec-3">3 Initialization   </h2>
 140 <div id="text-3">
 141
 142 <p><a href="/Users/kiwibird/Documents/Skole/V08/Probability/dmvccm/src/dmv.py">dmv-inits</a>
 143 </p>
 144 <p>
 145 We do have to go through the corpus, since the probabilities are based
 146 on how far away in the sentence arguments are from their heads.
 147 </p><ul>
 148 <li id="sec-3.1"><span class="todo">TODO</span> Separate initialization to another file?                      &nbsp;&nbsp;&nbsp;<span class="tag">PRETTIER</span><br/>
 149 (It's rather messy.)
 150 </li>
 151 <li id="sec-3.2"><span class="todo">TOGROK</span> CCM Initialization    <br/>
 152 P<sub>SPLIT</sub> used here&hellip; how, again?
 153 </li>
 154 <li id="sec-3.3"><span class="done">DONE</span> DMV Initialization probabilities<br/>
 155 (from initialization frequency)
 156 </li>
 157 <li id="sec-3.4"><span class="done">DONE</span> DMV Initialization frequencies    <br/>
 158 <span class="timestamp-kwd">CLOSED: </span> <span class="timestamp">2008-05-27 Tue 20:04</span><br/>
 159 <ul>
 160 <li id="sec-3.4.1">P_STOP    <br/>
 161 P<sub>STOP</sub> is not well defined by K&amp;M. One possible interpretation given
 162 the sentence [det nn vb nn] is
 163 <pre>
 164  f_{STOP}( STOP|det, L, adj) +1
 165  f_{STOP}(-STOP|det, L, adj) +0
 166  f_{STOP}( STOP|det, L, non_adj) +1
 167  f_{STOP}(-STOP|det, L, non_adj) +0
 168  f_{STOP}( STOP|det, R, adj) +0
 169  f_{STOP}(-STOP|det, R, adj) +1
 170
 171  f_{STOP}( STOP|nn, L, adj) +0
 172  f_{STOP}(-STOP|nn, L, adj) +1
 173  f_{STOP}( STOP|nn, L, non_adj) +1  # since there's at least one to the left
 174  f_{STOP}(-STOP|nn, L, non_adj) +0
 175 </pre>
 176 <ul>
 177 <li id="sec-3.4.1.1"><span class="todo">TODO</span> tweak<br/>
 178 <pre>
 179             f[head,  'STOP', 'LN'] += (i_h &lt;= 1)     # first two words
 180             f[head, '-STOP', 'LN'] += (not i_h &lt;= 1)
 181             f[head,  'STOP', 'LA'] += (i_h == 0)     # very first word
 182             f[head, '-STOP', 'LA'] += (not i_h == 0)
 183             f[head,  'STOP', 'RN'] += (i_h &gt;= n - 2) # last two words
 184             f[head, '-STOP', 'RN'] += (not i_h &gt;= n - 2)
 185             f[head,  'STOP', 'RA'] += (i_h == n - 1) # very last word
 186             f[head, '-STOP', 'RA'] += (not i_h == n - 1)
 187 </pre>
 188 vs
 189 <pre>
 190             # this one requires some additional rewriting since it
 191             # introduces divisions by zero
 192             f[head,  'STOP', 'LN'] += (i_h == 1)     # second word
 193             f[head, '-STOP', 'LN'] += (not i_h &lt;= 1) # not first two
 194             f[head,  'STOP', 'LA'] += (i_h == 0)     # first word
 195             f[head, '-STOP', 'LA'] += (not i_h == 0) # not first
 196             f[head,  'STOP', 'RN'] += (i_h == n - 2)     # second-to-last
 197             f[head, '-STOP', 'RN'] += (not i_h &gt;= n - 2) # not last two
 198             f[head,  'STOP', 'RA'] += (i_h == n - 1)     # last word
 199             f[head, '-STOP', 'RA'] += (not i_h == n - 1) # not last
 200 </pre>
 201 vs
 202 <pre>
 203             f[head,  'STOP', 'LN'] += (i_h == 1)     # second word
 204             f[head, '-STOP', 'LN'] += (not i_h == 1) # not second
 205             f[head,  'STOP', 'LA'] += (i_h == 0)     # first word
 206             f[head, '-STOP', 'LA'] += (not i_h == 0) # not first
 207             f[head,  'STOP', 'RN'] += (i_h == n - 2)     # second-to-last
 208             f[head, '-STOP', 'RN'] += (not i_h == n - 2) # not second-to-last
 209             f[head,  'STOP', 'RA'] += (i_h == n - 1)     # last word
 210             f[head, '-STOP', 'RA'] += (not i_h == n - 1) # not last
 211 </pre>
 212 vs
 213 "all words take the same number of arguments" interpreted as
 214 <pre>
 215 for all heads:
 216     p_STOP(head, 'STOP', 'LN') = 0.3
 217     p_STOP(head, 'STOP', 'LA') = 0.5
 218     p_STOP(head, 'STOP', 'RN') = 0.4
 219     p_STOP(head, 'STOP', 'RA') = 0.7
 220 </pre>
 221 (which we easily may tweak in init_zeros())
 222 </li>
 223 </ul>
 224 </li>
 225 <li id="sec-3.4.2">P_CHOOSE<br/>
 226 Go through the corpus, counting distances between heads and
 227 arguments. In [det nn vb nn], we give
 228 <ul>
 229 <li>
 230 f<sub>CHOOSE</sub>(nn|det, R) +1/1 + C
 231 </li>
 232 <li>
 233 f<sub>CHOOSE</sub>(vb|det, R) +1/2 + C
 234 </li>
 235 <li>
 236 f<sub>CHOOSE</sub>(nn|det, R) +1/3 + C
 237 <ul>
 238 <li>
 239 If this were the full corpus, P<sub>CHOOSE</sub>(nn|det, R) would have
 240 (1+1/3+2C) / sum_a f<sub>CHOOSE</sub>(a|det, R)
 241
 242 </li>
 243 </ul>
 244 </li>
 245 </ul>
 246
 247 <p>The ROOT gets "each argument with equal probability", so in a sentence
 248 of three words, 1/3 for each (in [nn vb nn], 'nn' gets 2/3). Basically
 249 just a frequency count of the corpus&hellip;
 250 </p></li>
 251 </ul>
 252 </li>
 253 </ul>
 254 </div>
 255
 256 </div>
 257
 258 <div id="outline-container-4" class="outline-2">
 259 <h2 id="sec-4">4 [#C] Deferred</h2>
 260 <div id="text-4">
 261
 262 <ul>
 263 <li id="sec-4.1"><span class="todo">TODO</span> if loc_h == t, no need to try right-attachment rules          &nbsp;&nbsp;&nbsp;<span class="tag">OPTIMIZE</span><br/>
 264 ** TODO if loc_h == s, no need to try left-attachment rules           :OPTIMIZE:
 265 </li>
 266 <li id="sec-4.3"><span class="done">DONE</span> inner_dmv() should disregard rules with heads not in sent     &nbsp;&nbsp;&nbsp;<span class="tag">OPTIMIZE</span><br/>
 267 <span class="timestamp-kwd">CLOSED: </span> <span class="timestamp">2008-06-08 Sun 10:18</span><br/>
 268 If the sentence is "nn vbd det nn", we should not even look at rules
 269 where
 270 <pre>
 271  rule.head() not in "nn vbd det nn".split()
 272 </pre>
 273 This is ruled out by getting rules from g.rules(LHS, sent).
 274
 275 <p>
 276 Also, we optimize this further by saying we don't even recurse into
 277 attachment rules where
 278 <pre>
 279  rule.head() not in sent[ s :r+1]
 280  rule.head() not in sent[r+1:t+1]
 281 </pre>
 282 meaning, if we're looking at the span "vbd det", we only use
 283 attachment rules where both daughters are members of ['vbd','det']
 284 (although we don't (yet) care about removing rules that rewrite to the
 285 same tag if there are no duplicate tags in the span, etc., that would
 286 be a lot of trouble for little potential gain).
 287 </p></li>
 288 <li id="sec-4.4"><span class="todo">TODO</span> when reestimating P_STOP etc, remove rules with p &lt; epsilon   &nbsp;&nbsp;&nbsp;<span class="tag">OPTIMIZE</span><br/>
 289 </li>
 290 <li id="sec-4.5"><span class="todo">TODO</span> inner_dmv, short ranges and impossible attachment             &nbsp;&nbsp;&nbsp;<span class="tag">OPTIMIZE</span><br/>
 291 If s-t &lt;= 2, there can be only one attachment below, so don't recurse
 292 with both Lattach=True and Rattach=True.
 293
 294 <p>
 295 If s-t &lt;= 1, there can be no attachment below, so only recurse with
 296 Lattach=False, Rattach=False.
 297 </p>
 298 <p>
 299 Put this in the loop under rewrite rules (could also do it in the STOP
 300 section, but that would only have an effect on very short sentences).
 301 </p></li>
 302 <li id="sec-4.6"><span class="todo">TODO</span> clean up the module files                                     &nbsp;&nbsp;&nbsp;<span class="tag">PRETTIER</span><br/>
 303 Is there better way to divide dmv and harmonic? There's a two-way
 304 dependency between the modules. Guess there could be a third file that
 305 imports both the initialization and the actual EM stuff, while a file
 306 containing constants and classes could be imported by all others:
 307 <pre>
 308  dmv.py imports dmv_EM.py imports dmv_classes.py
 309  dmv.py imports dmv_inits.py imports dmv_classes.py
 310 </pre>
 311
 312 </li>
 313 <li id="sec-4.7"><span class="todo">TOGROK</span> Some (tagged) sentences are bound to come twice             &nbsp;&nbsp;&nbsp;<span class="tag">OPTIMIZE</span><br/>
 314 Eg, first sort and count, so that the corpus
 315 [['nn','vbd','det','nn'],
 316 ['vbd','nn','det','nn'],
 317 ['nn','vbd','det','nn']]
 318 becomes
 319 [(['nn','vbd','det','nn'],2),
 320 (['vbd','nn','det','nn'],1)]
 321 and then in each loop through sentences, make sure we handle the
 322 frequency correctly.
 323
 324 <p>
 325 Is there much to gain here?
 326 </p>
 327 </li>
 328 <li id="sec-4.8"><span class="todo">TOGROK</span> tags as numbers or tags as strings?                         &nbsp;&nbsp;&nbsp;<span class="tag">OPTIMIZE</span><br/>
 329 Need to clean up the representation.
 330
 331 <p>
 332 Stick with tag-strings in initialization then switch to numbers for
 333 IO-algorithm perhaps? Can probably afford more string-matching in
 334 initialization..
 335 </p></li>
 336 </ul>
 337 </div>
 338
 339 </div>
 340
 341 <div id="outline-container-5" class="outline-2">
 342 <h2 id="sec-5">5 Adjacency and combining it with inner()</h2>
 343 <div id="text-5">
 344
 345 <p>Each DMV_Rule now has both a probN and a probA, for
 346 adjacencies. inner() needs the correct one in each case.
 347 </p>
 348 <p>
 349 Adjacency gives a problem with duplicate words/tags, eg. in the
 350 sentence "a a b". If this has the dependency structure b-&gt;a<sub>0</sub>-&gt;a<sub>1</sub>,
 351 then b is non-adjacent to a<sub>0</sub> and should use probN (for the LRStop and
 352 the attachment of a<sub>0</sub>), while the other rules should all use
 353 probA. But within the e(0,2,b) we can't just say "oh, a has index 0
 354 so it's not adjacent to 2", since there's also an a at index 1, and
 355 there's also a dependency structure b-&gt;a<sub>1</sub>-&gt;a<sub>0</sub> for that. We want
 356 both. And in possibly much more complex versions.
 357 </p>
 358 <p>
 359 So now we call inner() for each location of a head, and on each
 360 terminal, loc_h must equal s (and t). In the recursive attachment
 361 calls, we use the locations (sentence indices) of words to the left or
 362 right of the head in calls to inner(). loc_h lets us check whether we
 363 need probN or probA.
 364 </p>
 365 <p>
 366 In each inner() call, loc_h is the location of the root of this
 367 dependency structure.
 368 </p><ul>
 369 <li id="sec-5.1"><span class="done">DONE</span> [#A] test and debug my brilliant idea<br/>
 370 <span class="timestamp-kwd">CLOSED: </span> <span class="timestamp">2008-06-08 Sun 10:28</span><br/>
 371 </li>
 372 <li id="sec-5.2"><span class="done">DONE</span> implement my brilliant idea.<br/>
 373 <span class="timestamp-kwd">CLOSED: </span> <span class="timestamp">2008-06-01 Sun 17:19</span><br/>
 374 <a href="src/dmv.py">e(sti) in dmv.py</a>
 375
 376 </li>
 377 <li id="sec-5.3"><span class="done">DONE</span> [#A] test inner() on sentences with duplicate words<br/>
 378 Works with eg. the sentence "h h h"
 379 </li>
 380 </ul>
 381 </div>
 382
 383 </div>
 384
 385 <div id="outline-container-6" class="outline-2">
 386 <h2 id="sec-6">6 Expectation Maximation in IO/DMV-terms</h2>
 387 <div id="text-6">
 388
 389 <p>inner(s,t,LHS) calculates the expected number of trees headed by LHS
 390 from s to t (sentence positions). This uses the P_STOP and P_CHOOSE
 391 values, which have been conveniently distributed into CNF rules as
 392 probN and probA (non-adjacent and adjacent probabilites).
 393 </p>
 394 <p>
 395 When re-estimating, we use the expected values from inner() to get new
 396 values for P_STOP and P_CHOOSE. When we've re-estimated for the entire
 397 corpus, we distribute P_STOP and P_CHOOSE into the CNF rules again, so
 398 that in the next round we use new probN and probA to find
 399 inner-probabilites.
 400 </p>
 401 <p>
 402 The distribution of P_STOP and P_CHOOSE into CNF rules also happens in
 403 init_normalize() (here along with the creation of P_STOP and
 404 P_CHOOSE); P_STOP is used to create CNF rules where one branch of the
 405 rule is STOP, P_CHOOSE is used to create rules of the form
 406 <pre>
 407  h  -&gt; h  _a_
 408  h_ -&gt; h_ _a_
 409 </pre>
 410 </p>
 411 <p>
 412 Since "adjacency" is not captured in regular CNF rules, we need two
 413 probabilites for each rule, and inner() has to know when to use which.
 414 </p>
 415 <ul>
 416 <li id="sec-6.1"><span class="todo">TODO</span> Corpus access<br/>
 417 </li>
 418 <li id="sec-6.2"><span class="todo">TOGROK</span> sentences or rules as the "outer loop"?                     &nbsp;&nbsp;&nbsp;<span class="tag">OPTIMIZE</span><br/>
 419 In regard to the E/M-step, finding P<sub>STOP</sub>, P<sub>CHOOSE</sub>.
 420
 421
 422 </li>
 423 </ul>
 424 </div>
 425
 426 </div>
 427
 428 <div id="outline-container-7" class="outline-2">
 429 <h2 id="sec-7">7 Python-stuff</h2>
 430 <div id="text-7">
 431
 432 <ul>
 433 <li>
 434 <a href="src/pseudo.py">pseudo.py</a>
 435 </li>
 436 <li>
 437 <a href="http://nltk.org/doc/en/structured-programming.html">http://nltk.org/doc/en/structured-programming.html</a> recursive dynamic
 438 </li>
 439 <li>
 440 <a href="http://nltk.org/doc/en/advanced-parsing.html">http://nltk.org/doc/en/advanced-parsing.html</a>
 441 </li>
 442 <li>
 443 <a href="http://jaynes.colorado.edu/PythonIdioms.html">http://jaynes.colorado.edu/PythonIdioms.html</a>
 444
 445
 446
 447 </li>
 448 </ul>
 449 </div>
 450
 451 </div>
 452
 453 <div id="outline-container-8" class="outline-2">
 454 <h2 id="sec-8">8 Git</h2>
 455 <div id="text-8">
 456
 457 <p>Setting up a new project:
 458 <pre>
 459  git init
 460  git add .
 461  git commit -m "first release"
 462 </pre>
 463 </p>
 464 <p>
 465 Later on: (<code>-a</code> does <code>git rm</code> and <code>git add</code> automatically)
 466 <pre>
 467  git init
 468  git commit -a -m "some subsequent release"
 469 </pre>
 470 </p>
 471 <p>
 472 Then push stuff up to the remote server:
 473 <pre>
 474  git push git+ssh://username@repo.or.cz/srv/git/dmvccm.git master
 475 </pre>
 476 </p>
 477 <p>
 478 (<code>eval `ssh-agent`</code> and <code>ssh-add</code> to avoid having to type in keyphrase all
 479 the time)
 480 </p>
 481 <p>
 482 Make a copy of the (remote) master branch:
 483 <pre>
 484  git clone git://repo.or.cz/dmvccm.git
 485 </pre>
 486 </p>
 487 <p>
 488 Make and name a new branch in this folder
 489 <pre>
 490  git checkout -b mybranch
 491 </pre>
 492 </p>
 493 <p>
 494 To save changes in <code>mybranch</code>:
 495 <pre>
 496  git commit -a
 497 </pre>
 498 </p>
 499 <p>
 500 Go back to the master branch (uncommitted changes from <code>mybranch</code> are
 501 carried over):
 502 <pre>
 503  git checkout master
 504 </pre>
 505 </p>
 506 <p>
 507 Try out:
 508 <pre>
 509  git add --interactive
 510 </pre>
 511 </p>
 512 <p>
 513 Good tutorial:
 514 <a href="http://www-cs-students.stanford.edu/~blynn//gitmagic/">http://www-cs-students.stanford.edu/~blynn//gitmagic/</a>
 515 </p></div>
 516 </div>
 517 <div id="postamble"><p class="author"> Author: Kevin Brubeck Unhammer
 518 <a href="mailto:K.BrubeckUnhammer at student uva nl ">&lt;K.BrubeckUnhammer at student uva nl &gt;</a>
 519 </p>
 520 <p class="date"> Date: 2008/06/08 12:43:26</p>
 521 </div><p class="postamble">Skrive vha. emacs + <a href='http://orgmode.org/'>org-mode</a></p></body>
 522 </html>